メニュー
Guide for creating effective skills following best practices. Use when creating or updating skills that extend agent capabilities.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | create-skill |
| description | Guide for creating effective skills following best practices. Use when creating or updating skills that extend agent capabilities. |
Use /create-skill to create or update effective skills that extend agent capabilities with specialized knowledge, workflows, and tool integrations.
Skills are modular, self-contained packages that extend agent capabilities by providing specialized knowledge, workflows, and tools. Think of them as "onboarding guides" for specific domains or tasks.
The 200-line rule is critical. SKILL.md must be under 200 lines. If you need more, split content into references/ files.
references/)skill-name/
├── SKILL.md (required, <200 lines)
│ ├── YAML frontmatter metadata (required)
│ │ ├── name: (required)
│ │ └── description: (required)
│ └── Markdown instructions (required)
└── Bundled Resources (optional)
├── scripts/ - Executable code
├── references/ - Documentation loaded as needed
└── assets/ - Files used in output
The context window is a shared resource. Your skill shares it with everything else the agent needs. Be concise and challenge each piece of information:
Skills act as additions to models, so effectiveness depends on the underlying model. Test your skill with all models you plan to use it with.
For detailed guidance, see:
references/progressive-disclosure.md - 200-line rule and references patternreferences/skill-structure.md - SKILL.md format and frontmatter detailsreferences/examples.md - Good skill examplesreferences/best-practices.md - Comprehensive best practices guideUse when a user wants to compare an arbitrary skill or path across models, quality, cost, latency, or routing.
Analyze actual local Copilot skill activations without reading conversation content.
Plan work too big for one agent session as a shared map of decision tickets on GitHub Issues or Jira, then resolve them one at a time until the route is clear. Use with /wayfinder when the user says "plan something too big for one session", "chart a map", "wayfind", "work the map", or "decision tickets". Always confirm tracker (GitHub vs Jira) and destination before writing. Adapted from mattpocock/skills wayfinder (MIT).
Create or resume a task handoff so a fresh agent can continue unfinished work. Persist via omp handoff CLI (never write handoff files yourself). Use when the user says /handoff, wants to pause mid-task, or resume prior work.
Single-owner execute-fix-verify loop for one clear task. Use with /ralph when one agent should keep going until evidence or blocker.
Adversarial QA pass that tests behavior, failures, and regressions. Use with /ultraqa after implementation when shallow checks are not enough.